Recently we discovered a serious bug in the compiler whereby we miscompiled several packages. The problem started with my ABI-changing patch which changed how empty classes are passed, as per the x86_64 psABI (so this bug only affects x86_64). The problem could arise when the code contained empty class templates. For more info see https://gcc.gnu.org/PR84502.
I did another mass rebuild with a specially-tweaked gcc in order to find out which packages need to be rebuild with patched gcc-8.0.1-0.16. Sorry about this.
This is the list:
0ad-0.0.22-5.fc28.src.rpm adonthell-0.3.6-7.fc28.src.rpm american-fuzzy-lop-2.52b-2.fc28.src.rpm ardour4-4.7.0-10.fc28.src.rpm ardour5-5.12.0-4.fc28.src.rpm BEDTools-2.26.0-3.fc26.src.rpm bro-2.5.3-1.fc28.src.rpm bullet-2.87-2.fc28.src.rpm camotics-1.1.1-11.fc28.src.rpm Canna-3.7p3-52.fc28.src.rpm crawl-0.21.1-2.fc28.src.rpm dx-4.4.4-44.fc28.src.rpm dyninst-9.3.2-10.fc28.src.rpm eiskaltdcpp-2.2.11-12.20180207git6ca065b.fc28.src.rpm enblend-4.2-9.fc28.src.rpm endless-sky-0.9.8-4.fc28.src.rpm eureka-1.00-11.fc28.src.rpm fastbit-2.0.3-5.fc28.src.rpm fbreader-0.12.10-18.fc23.src.rpm fcl-0.5.0-8.fc28.src.rpm freecad-0.16-11.fc28.src.rpm freenx-server-0.7.3-41.fc28.src.rpm freeorion-0.4.7.1-6.fc28.src.rpm glslang-3.1-0.6.20180205.git2651cca.fc28.src.rpm gnucap-0.35-24.fc28.src.rpm gtk4-3.92.1-2.fc28.src.rpm hypre-2.13.0-5.fc28.src.rpm ibp-0.21-17.fc28.src.rpm isdn4k-utils-3.27-10.fc28.src.rpm java-1.8.0-openjdk-1.8.0.161-8.b14.fc28.src.rpm kf5-libkleo-17.12.2-1.fc28.src.rpm kinput2-v3.1-58.fc28.src.rpm koules-1.4-25.fc28.src.rpm kyotocabinet-1.2.76-16.fc28.src.rpm lcgdm-1.10.0-1.fc28.src.rpm libimagequant-2.11.7-2.fc28.src.rpm libqb-1.0.3-2.fc28.src.rpm librime-1.2-21.fc28.src.rpm llvm34-3.4.2-10.fc26.src.rpm llvm35-3.5.2-4.fc26.src.rpm llvm3.7-3.7.1-7.fc27.src.rpm llvm3.9-3.9.1-11.fc27.src.rpm llvm4.0-4.0.1-3.fc28.src.rpm llvm5.0-5.0.1-4.fc28.src.rpm MagicPoint-1.13a-20.fc28.src.rpm mld2p4-2.1.1-0.4.fc28.src.rpm mlpack-2.2.5-3.fc28.src.rpm mmapper-2.4.5-1.fc28.src.rpm nas-1.9.4-11.fc28.src.rpm nco-4.7.1-1.fc28.src.rpm netpanzer-0.8.7-3.fc26.src.rpm nx-libs-3.5.0.33-3.fc28.src.rpm octave-image-2.6.2-1.fc28.src.rpm oggvideotools-0.9-5.fc27.src.rpm oneko-1.2-24.fc28.src.rpm openms-2.3.0-7.fc28.src.rpm openmsx-0.14.0-2.fc28.src.rpm p7zip-16.02-10.fc28.src.rpm pgRouting-2.5.2-3.fc28.src.rpm pngquant-2.11.7-3.fc28.src.rpm polyclipping-6.4.2-2.fc28.src.rpm pythia8-8.2.15-6.fc28.src.rpm python-pyclipper-1.1.0-1.fc28.src.rpm qt5-qtlocation-5.10.1-1.fc28.src.rpm quantum-espresso-5.4.0-12.fc26.src.rpm rasmol-2.7.5.2-8.fc27.src.rpm renderdoc-0.91-6.fc28.src.rpm rocksdb-5.7.3-2.fc28.src.rpm root-tail-1.2-20.fc28.src.rpm rosegarden4-17.12-3.fc28.src.rpm sendmail-8.15.2-23.fc28.src.rpm seqan-1.4.2-34.fc28.src.rpm seqan2-2.4.0-2.fc28.src.rpm stage-4.1.1-17.fc27.src.rpm stxxl-1.4.1-6.fc28.src.rpm supertuxkart-0.9.3-2.fc28.2.src.rpm synergy-2.0.0-2.fc28.src.rpm tapkee-1.1-6.fc28.src.rpm tcpxtract-1.0.1-26.fc28.src.rpm tgif-4.2.5-15.fc28.src.rpm unclutter-8-16.fc28.src.rpm v8-6.2.91-5.fc28.src.rpm vigra-1.11.1-4.fc28.src.rpm vtk-7.1.1-10.fc28.src.rpm vxl-1.17.0-25.fc28.src.rpm x11-ssh-askpass-1.2.4.1-21.fc28.src.rpm xapian-core-1.4.5-2.fc28.src.rpm xautolock-2.2-18.fc24.src.rpm xkeycaps-2.46-22.fc28.src.rpm xmountains-2.9-2.D20170103git3ba444a4f7.fc28.1.src.rpm xskat-4.0.0-19.fc28.src.rpm xstar-2.2.0-18.fc28.src.rpm xvkbd-3.7-6.fc28.src.rpm yadex-1.7.0-45.fc28.src.rpm
The following packages were affected, but have already been rebuilt:
cbmc-5.7-3.fc27.src.rpm clamav-0.99.3-7.fc28.src.rpm ceph-12.2.2-1.fc28.src.rpm chromium-64.0.3282.119-1.fc28.src.rpm glibc-2.27-3.fc28.src.rpm golang-1.10-0.rc2.1.fc28.src.rpm libepoxy-1.4.3-6.fc28.src.rpm librealsense-2.10.0-1.fc28.src.rpm mapnik-3.0.18-1.fc28.src.rpm mongodb-3.6.2-5.fc28.src.rpm opencv-3.3.1-4.fc28.src.rpm opengrm-ngram-1.3.2-7.fc27.src.rpm pcl-1.8.1-1.fc28.src.rpm qbittorrent-4.0.3-3.fc28.src.rpm root-6.12.06-1.fc28.src.rpm sphinxtrain-1.0.8-39.fc28.src.rpm
These two packages couldn't be rebuild because of lack of memory. I think they should be rebuilt too, for a good measure:
openvswitch-2.8.1-2.fc28.src.rpm openblas-0.2.20-6.fc28.src.rpm
Thanks,
Marek
On Wed, 2018-03-07 at 19:50 +0100, Marek Polacek wrote:
I did another mass rebuild with a specially-tweaked gcc in order to find out which packages need to be rebuild with patched gcc-8.0.1-0.16.
How we find out the same problem in packages from external repos ?
Thanks,
On Wed, Mar 07, 2018 at 07:20:18PM +0000, Sérgio Basto wrote:
On Wed, 2018-03-07 at 19:50 +0100, Marek Polacek wrote:
I did another mass rebuild with a specially-tweaked gcc in order to find out which packages need to be rebuild with patched gcc-8.0.1-0.16.
How we find out the same problem in packages from external repos ?
I don't know; I've only rebuilt all rawhide packages. Were the external packages built with gcc 8? If not, we're fine.
Marek
On Thu, 2018-03-08 at 10:25 +0100, Marek Polacek wrote:
On Wed, Mar 07, 2018 at 07:20:18PM +0000, Sérgio Basto wrote:
On Wed, 2018-03-07 at 19:50 +0100, Marek Polacek wrote:
I did another mass rebuild with a specially-tweaked gcc in order to find out which packages need to be rebuild with patched gcc-8.0.1-0.16.
How we find out the same problem in packages from external repos ?
I don't know; I've only rebuilt all rawhide packages.
You rebuilt all packages and found what packages was different ? that'sit ?
Were the external packages built with gcc 8?
RPMFusion
If not, we're fine.
Marek _______________________________________________ devel mailing list -- devel@lists.fedoraproject.org To unsubscribe send an email to devel-leave@lists.fedoraproject.org
On Thu, Mar 08, 2018 at 05:52:05PM +0000, Sérgio Basto wrote:
On Thu, 2018-03-08 at 10:25 +0100, Marek Polacek wrote:
On Wed, Mar 07, 2018 at 07:20:18PM +0000, Sérgio Basto wrote:
On Wed, 2018-03-07 at 19:50 +0100, Marek Polacek wrote:
I did another mass rebuild with a specially-tweaked gcc in order to find out which packages need to be rebuild with patched gcc-8.0.1-0.16.
How we find out the same problem in packages from external repos ?
I don't know; I've only rebuilt all rawhide packages.
You rebuilt all packages and found what packages was different ? that'sit ?
The bug affected solely x86_64 and the test mass rebuilt was thus performed solely on x86_64, on non-koji boxes (even when the compiler for checking this was twice as slow as normally). Rebuilding all packages again on all architectures would be certainly much more CPU time.
Jakub
On Wed, 2018-03-07 at 19:50 +0100, Marek Polacek wrote:
Recently we discovered a serious bug in the compiler whereby we miscompiled several packages. The problem started with my ABI-changing patch which changed how empty classes are passed, as per the x86_64 psABI (so this bug only affects x86_64). The problem could arise when the code contained empty class templates. For more info see https://gcc.gnu.org/PR84502.
I did another mass rebuild with a specially-tweaked gcc in order to find out which packages need to be rebuild with patched gcc-8.0.1-0.16. Sorry about this.
This is the list:
<snip>
xautolock-2.2-18.fc24.src.rpm
This seems like an odd entry. How can a package last built for F24 possibly be affected?
Other than that...this is a bit awkward as F28 is now in freeze. I will try and rebuild all the affected packages and submit an update and an FE bug.
On Wed, Mar 07, 2018 at 12:27:34PM -0800, Adam Williamson wrote:
On Wed, 2018-03-07 at 19:50 +0100, Marek Polacek wrote:
Recently we discovered a serious bug in the compiler whereby we miscompiled several packages. The problem started with my ABI-changing patch which changed how empty classes are passed, as per the x86_64 psABI (so this bug only affects x86_64). The problem could arise when the code contained empty class templates. For more info see https://gcc.gnu.org/PR84502.
I did another mass rebuild with a specially-tweaked gcc in order to find out which packages need to be rebuild with patched gcc-8.0.1-0.16. Sorry about this.
This is the list:
<snip>
xautolock-2.2-18.fc24.src.rpm
This seems like an odd entry. How can a package last built for F24 possibly be affected?
Just guessing; Marek has rebuilt all the non-noarch src.rpm for rawhide and the package build diagnosed the ABI incompatibility. Perhaps the build normally only fails later than where the ABI issue was spotted.
The instrumented GCC had a new option to compile using the previous (8.0.1-0.15 and earlier) wrong behavior and compiled everything twice, comparing dumps on what would be produced between the two.
Jakub
On Wed, Mar 07, 2018 at 09:36:29PM +0100, Jakub Jelinek wrote:
On Wed, Mar 07, 2018 at 12:27:34PM -0800, Adam Williamson wrote:
On Wed, 2018-03-07 at 19:50 +0100, Marek Polacek wrote:
Recently we discovered a serious bug in the compiler whereby we miscompiled several packages. The problem started with my ABI-changing patch which changed how empty classes are passed, as per the x86_64 psABI (so this bug only affects x86_64). The problem could arise when the code contained empty class templates. For more info see https://gcc.gnu.org/PR84502.
I did another mass rebuild with a specially-tweaked gcc in order to find out which packages need to be rebuild with patched gcc-8.0.1-0.16. Sorry about this.
This is the list:
<snip>
xautolock-2.2-18.fc24.src.rpm
This seems like an odd entry. How can a package last built for F24 possibly be affected?
Just guessing; Marek has rebuilt all the non-noarch src.rpm for rawhide
Exactly.
and the package build diagnosed the ABI incompatibility. Perhaps the build normally only fails later than where the ABI issue was spotted.
The instrumented GCC had a new option to compile using the previous (8.0.1-0.15 and earlier) wrong behavior and compiled everything twice, comparing dumps on what would be produced between the two.
Of course packages that were never built with gcc 8 are not affected, and need not be rebuilt, but the list doesn't reflect that.
Marek
On Wed, 2018-03-07 at 19:50 +0100, Marek Polacek wrote:
Recently we discovered a serious bug in the compiler whereby we miscompiled several packages. The problem started with my ABI-changing patch which changed how empty classes are passed, as per the x86_64 psABI (so this bug only affects x86_64). The problem could arise when the code contained empty class templates. For more info see https://gcc.gnu.org/PR84502.
I did another mass rebuild with a specially-tweaked gcc in order to find out which packages need to be rebuild with patched gcc-8.0.1-0.16. Sorry about this.
This is the list:
I've fired rebuilds of every package on the list, for F28 and Rawhide. I'll investigate any failures and put together an F28 update tomorrow.
On Wed, Mar 07, 2018 at 09:19:15PM -0800, Adam Williamson wrote:
On Wed, 2018-03-07 at 19:50 +0100, Marek Polacek wrote:
Recently we discovered a serious bug in the compiler whereby we miscompiled several packages. The problem started with my ABI-changing patch which changed how empty classes are passed, as per the x86_64 psABI (so this bug only affects x86_64). The problem could arise when the code contained empty class templates. For more info see https://gcc.gnu.org/PR84502.
I did another mass rebuild with a specially-tweaked gcc in order to find out which packages need to be rebuild with patched gcc-8.0.1-0.16. Sorry about this.
This is the list:
I've fired rebuilds of every package on the list, for F28 and Rawhide. I'll investigate any failures and put together an F28 update tomorrow.
Excellent. I saw many failures due to "missing package". Otherwise you can take a look at https://lists.fedoraproject.org/archives/list/devel@lists.fedoraproject.org/... to see if some failures persist.
Marek
-
Adam Williamson -
Jakub Jelinek -
Marek Polacek -
Sérgio Basto